I happen to know those intersections. They could have improved that wiggle route simply by rating the intersections. I'm not sure but it looks like they were not doing that.

The old route crossed two stop signs and did a left turn at a traffic signal. Left turns at traffic signals really suck.

The new, proper wiggle route is two left turns and a right turn, all at stop signs. Left turns at stop signs are not that much worse than going straight across -- and vastly better than a traffic light. Right turns are practically free.

So by simply rating the difficulty of intersections using a global metric, it seems like they could have made the same wiggle improvement without getting into rating each section of street. My point here is not that assessing things like accomodation_factor and road_stress are useless. Rather it appears that they've overlooked the much simpler, global improvement of rating intersections.